CSAIL Research Abstracts - 2005 link to http://publications.csail.mit.edu/abstracts/abstracts05/index.html link to http://www.csail.mit.edu
bullet Introduction bullet Architecture, Systems
& Networks
bullet Language, Learning,
Vision & Graphics
bullet Physical, Biological
& Social Systems
bullet Theory bullet

horizontal line

Nonlinear Precomputed Radiance Transfer

Paul Green, Jan Kautz, Wojciech Matusik, Fredo Durand & Henrik Wann Jensen

brief image
      description goes here

Figure 1: A highly glossy object illuminated under two different environments. Notice the inter-reflections of the tail on the head (left image) and foot (right image).


We propose a real-time method for rendering static objects with complex materials under distant all-frequency lighting. Existing precomputed light transport approaches [1,2] can render objects with complex material properties (e.g., anisotropy, translucency) and complex global illumination effects (e.g., inter-reflections, soft shadows, subsurface scattering). However, they are either restricted to low-frequency illumination or they assume a fixed viewer. We propose a solution that allows both. Our method is based on two components: (1) a compact non-linear representation for precomputed light transport that can be integrated rapidly with the distant all-frequency illumination, and (2) a new interpolation scheme that enables rendering of an arbitrary viewpoint from only a sparse set of precomputed views. Figure 1 shows an object rendered using our method.


We can view the scene as a "black box" linear system that transforms the input radiance into the output radiance. This linear system can be written as:

where x is a point on the surface, wo is the outgoing direction, wi is the incoming direction, Li is the vector of (distant) incident radiance, T() is the light transport kernel (represented as a vector, with a coordinate for each direction wi), and Lo is the exitant radiance [Ng et al. 2003]. Both incident radiance Li and the transport kernel T() are sampled over the sphere of incoming directions. T() captures the complex transport effects occurring in the scene. We represent the transport vector with a non-linear approximation. T() is represented as a sum of constant-valued box functions (1 inside the box and 0 outside the box) of arbitrary size and position:

where Kj defines the size and position of box j and wj is a scalar weight of the box. This representation has two advantages: an efficient integration algorithm and a simple interpolation technique that allows us to approximate T(x,w'o ) for an arbitrary (i.e. not precomputed) direction w'o. Using this representation, we can approximate Equation 1 efficiently:

Using a summed area table representation for Li, we can compute < Kj, Li> quickly. For an arbitrary outgoing direction, w0o , it is not sufficient to interpolate the resulting Lo from the nearest precomputed view directions because of possible high-frequency viewdependent components (e.g., specular reflections). Instead, we synthesize a new light transport vector T(x,w'o ) by interpolating the box parameters (Kj and wj) from the three nearest precomputed directions. The synthesized transport vector is a barycentric combination of the three nearest transport vectors. T(x,w'o ) is then integrated with the incident lighting (Equation 3) to produce the exitant radiance Lo(x,w'o ) at position x along direction w'o .

Research Support

This research was supported by a Ford Foundation Graduate Fellowship, MIT Presidential Fellowship, NSF CAREER award 0447561 "Transient Signal Processing for Realistic Imagery," NSF CISE Research Infrastructure Award (EIA9802220), INRIA equipe associee and the MIT-France Program.


[1] R. Ng R. Ramamoorthi and P. Hanrahan. All-frequency shadows using non-linear wavelet lighting approximation. In ACM Transactions on Graphics 22,3 (July) pp. 376-381.

[2] P Sloan, J. Kautz and J. Snyder. Precomputed radiance transfer for real-time rendering in dynamic, low-frewuency lighting environments. In ACM Transactions on Graphics 21,3 (July) pp. 527-536.

horizontal line

MIT logo Computer Science and Artificial Intelligence Laboratory (CSAIL)
The Stata Center, Building 32 - 32 Vassar Street - Cambridge, MA 02139 - USA
tel:+1-617-253-0073 - publications@csail.mit.edu
(Note: On July 1, 2003, the AI Lab and LCS merged to form CSAIL.)